A Prototype System Using Causal Inference and Machine Learning Models to Estimate ESG Ratings for SMEs

Although numerous environmental, social, and governance (ESG) rating agencies exist globally, different organizations may employ distinct evaluation frameworks and methods, potentially leading to inconsistent or misleading ratings for a given company. Moreover, these rating agencies primarily focus on evaluating publicly listed large-cap companies, rendering their assessment approaches unsuitable for small and medium-sized enterprises (SMEs) that lack sufficient resources to support ESG evaluations. This study aims to develop an SME-specific ESG rating system that integrates both indirect and direct approaches. To facilitate responsible and reliable ESG evaluations, the indirect approach employs causal inference technique to identify the most relevant variables affecting ESG ratings in large listed companies. Subsequently, three machine learning-based predictive models are trained using this data, which are then applied to SMEs with a bridging strategy to derive their potential ESG ratings. The direct approach categorizes SMEs by industry and applies customized ESG performance indicators to assess their actual ESG ratings. Finally, the two rating results are integrated using a weighted formula, accompanied by the implementation of Microsoft Power BI's visualized and interactive features, to establish a transparent, explainable, and interpretable SME-specific ESG rating system.